Searcharxiv⌕ Search

arXiv subjects

Zhengzhong Tu

Publications and source records attributed to Zhengzhong Tu.

At least 19 recordsLinked to original sources
arXiv2026

Flex-RSI: Benchmarking Recursive Self-Improvement of Frontier Agents

Frontier language models are increasingly deployed as agents, and how well an agent performs depends not only on its model but also on its harness: the prompts, tools, memory, and control loop through which the model acts. When an agent is applied to a new problem, a natural question is whether it can use the feedback of the task to improve the way it works, and not merely its answer. Most existing benchmarks keep the agent's harness fixed and score the solution it produces, or target large-scale model training. We introduce Flex-RSI, a benchmark for the recursive self-improvement of agents on vision and robotics tasks under a fixed budget. In each round, an executor improves the task solution, which is kept only if it scores higher on held-out validation data, and an improver then revises the harness, including the executor and itself. Every run starts from the same general-purpose harness, and the final solution is scored on hidden test data. Evaluating closed-source and open-weight models on tasks spanning perception, reasoning, robot action, 3D reconstruction, visual creation, and infrastructure, we find that self-improvement helps most when the agent builds its own solution, that feedback recording only success or failure holds back open-weight models, and that open-weight models fall furthest behind where outputs must pass strict checks of physical stability or exact correctness. Paired runs further show that editing the harness itself contributes to the gains.

cs.AI↗
arXiv2026

FlashDiff: Efficient Regional Execution and Scheduling for Diffusion Model Serving

Diffusion models have become the central backbone for modern image, video, and audio generation, but their efficient service remains a challenge. Unlike autoregressive decoding, diffusion inference repeatedly updates high-dimensional spatial or temporal latents over many denoising steps. This all-region execution pattern makes generation latency high and limits serving throughput. Existing multi-GPU parallelization methods can reduce per-step computation, but often introduce substantial activation exchange overhead, causing communication to offset or even outweigh the benefits of parallel execution. This paper presents FlashDiff, a diffusion serving system that improves inference efficiency through adaptive regional execution and scheduling. FlashDiff is based on the observation that diffusion refinement is not uniform across latent regions or denoising steps: different regions often stabilize at different rates, while neighboring steps exhibit strong temporal correlation. FlashDiff leverages these properties to selectively execute only regions that require further refinement and to reallocate the resulting compute slack across concurrent serving requests. FlashDiff consists of three mechanisms. First, it decomposes the latent representation into coherent execution regions using early-stage attention signals, preserving semantic structure while exposing fine-grained parallelism. Second, it uses a lightweight runtime controller to estimate region activity and bypass low-impact updates when further refinement is unlikely to affect output quality. Third, it applies an affinity-aware online scheduler that co-locates dependent regions, balances residual load across GPUs, and reuses reclaimed compute capacity to improve serving efficiency. Across real-world image, video, and audio workloads, FlashDiff reduces end-to-end serving latency by 30-97% and improves throughput by 1.2-2.2x.

cs.DC↗
arXiv2026

From Self-Confidence to Statistical Accuracy: Scaling Human-in-the-Loop Evaluation with Direct and LLM-Assisted Auditing

For large language models (LLMs) and multimodal large language models (MLLMs), model-reported self-confidence is often miscalibrated, yet it may still be useful for efficient evaluation. At the same time, LLM-as-a-judge offers a scalable alternative to direct human auditing, but its practical benefit depends on calibration quality and inference cost. This paper develops a unified statistical framework for both evaluation routes under alternative sampling strategies. For direct auditing, confidence-stratified sampling is used to recover population accuracy from partially verified outputs without requiring calibrated confidence. Finite-sample uncertainty is characterized, and the analysis shows how highly uneven confidence strata affect audit efficiency and label allocation. For judge-assisted evaluation, the judge is modeled through sensitivity and false-positive rate, yielding an error-corrected estimator with calibration uncertainty. The two routes are then compared through human-label and total-cost efficiency boundaries. Experiments on MMLU-Pro and MMMU-Pro show that miscalibrated self-confidence can still support statistically valid stratified auditing. Under highly imbalanced confidence distributions, proportional allocation substantially reduces estimation error relative to equal-per-stratum allocation. For judge-assisted evaluation, only a small subset of target--verifier pairs satisfies the idealized human-label efficiency condition, and automated inference costs can further limit potential savings. These findings clarify the statistical and economic conditions under which confidence-guided auditing and automated verification are useful.

cs.CL↗
arXiv2026

FastVLA: Task-Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision-Language-Action Models

Vision-Language-Action (VLA) models present a promising paradigm for general-purpose robot manipulation, but closed-loop deployment is often limited by inference efficiency: at each timestep, high-resolution, often multi-view images are encoded into hundreds of visual tokens that dominate transformer computation and latency. Token compression via attention is a natural solution but is challenging for VLA control because (1) decision-critical tokens are determined by the task, not generic visual saliency, and (2) the token budget varies substantially across steps, so static metrics and budgets waste computation or discard control-critical tokens. We introduce FastVLA, a training-free, task-aware, and temporally adaptive visual token compression framework. At each step, FastVLA scores token importance via task-to-vision attention and adaptively sets the compression rate from the importance-distribution uniformity, measured by an inverse-Simpson metric with online sliding-window calibration. We evaluate FastVLA on OpenVLA, OpenVLA-OFT, and CogACT across LIBERO and SIMPLER, and on a physical Franka Research 3 robot. In simulation, it reduces LLM-backbone FLOPs by 35-59\% and action latency by 9-32\% (a 1.09$\times$-1.46$\times$ speedup) while maintaining task success, achieving a comparable or better efficiency-performance trade-off than existing plug-and-play token compression/pruning/reuse baselines. On the real robot, it cuts action-prediction latency by 20\% (1.25$\times$ speedup) at comparable success. Code is available at https://github.com/jiuyixu25/FastVLA.

cs.RO↗
arXiv2026

Are We Really Measuring Forgetting? Pitfalls in Evaluating Machine Unlearning for Multimodal LLMs

As machine unlearning for multimodal large language models (MLLMs) rapidly advances, rigorous evaluation of its effectiveness becomes increasingly important. However, existing benchmarks largely inherit evaluation protocols from text-only unlearning and overlook the unique challenges of unlearning in multimodal settings. In this work, we identify three major pitfalls that may lead to misleading conclusions: the lack of heldout evaluation to distinguish genuine concept-wise forgetting from instance-wise overfitting, the reliance on synthetic, fine-tuned knowledge rather than real-world knowledge acquired during pre-training, and unfair method comparisons caused by applying the same training settings to different unlearning methods. To address these challenges, we introduce GEMMU (Generalizable Evaluation for Multimodal Machine Unlearning), a benchmark built on real-world knowledge and a principled evaluation protocol. Re-evaluating representative unlearning methods under GEMMU reveals substantial generalization gaps and challenges conclusions drawn from previous benchmarks. Our findings highlight the importance of rigorous evaluation for advancing multimodal machine unlearning and provide guidelines for future benchmark design.

cs.CL↗
arXiv2026

LLMs Can Get "Brain Rot": A Pilot Study on Twitter/X

We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs). To unveil junk effects, we designed a novel controlled experiment on real Twitter/X corpora, by constructing junk and reverse-controlled datasets via two orthogonal operationalizations: M1 (engagement degree) and M2 (semantic quality), with matched token scale and training operations across conditions. Compared to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism). The gradual mixtures of junk and control datasets also yield dose-response cognition decay: for example, under M1, ARC-Challenge with Chain-of-Thought drops 72.1 -> 57.2 and RULER-CWE 83.7 -> 52.3 as junk ratio rises from 0% to 100%. Error forensics reveal several key insights. First, we identify thought-skipping as the primary lesion in reasoning: models increasingly truncate or skip chains. Second, partial but incomplete healing is observed: scaling instruction tuning and clean continual pre-training improve the declined cognition, yet cannot restore baseline capability, suggesting persistent representational drift rather than format mismatch. Finally, we discover that the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that social effects of data could be a causal driver of LLM capability decay in continual pre-training, thereby motivating routine "cognitive health checks" for deployed and evolving LLMs.

cs.CL↗
arXiv2026

AdaptFuse: Training-Free Sequential Preference Inference via Externalized Bayesian Computation

Large language models (LLMs) struggle to accumulate evidence across multiple rounds of user interactions, failing to update their beliefs in a manner consistent with Bayesian inference. Existing solutions require fine-tuning on sensitive user interaction data, limiting their applicability in privacy-conscious settings. We propose \textsc{AdaptFuse}, a training-free framework that externalizes probabilistic computation entirely from the LLM, where a symbolic module maintains a Bayesian posterior over a discrete hypothesis set, while a frozen LLM contributes semantic reasoning via multi-sample Dirichlet aggregation. The two signals are combined through entropy-adaptive fusion, which automatically weights each source by its predictive confidence, shifting reliance from the LLM to the symbolic posterior as evidence accumulates. We evaluate across three domains: flight recommendation, hotel recommendation, and web shopping; on Gemma~2 9B, Llama~3 8B, and Qwen~2.5 7B. \textsc{AdaptFuse} consistently outperforms both prompting baselines and fine-tuned Bayesian Teaching models on all tasks, with accuracy improving monotonically over interaction rounds. These results demonstrate that principled inference-time algorithms can substitute for fine-tuning in personalized recommendation, without storing or training on sensitive user data.

cs.CL↗
arXiv2026

PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning

Large Reasoning Language Models (LRMs) tackle complex tasks through test-time scaling, generating long-form Chain-of-Thought (CoT) trajectories during inference. These trajectories contain reflection markers such as wait, but, and alternatively, accompanying hesitation, revision, and alternative explorations. Recent test-time control methods use these markers to steer reasoning, but treating them as one coarse-grained category can obscure their distinct intervention effects. We study how marker identity and reasoning context shape these effects. Type-wise suppression produces different accuracy--length tradeoffs, while fixed-prefix interventions show that the relative benefit of continuation and revision varies across prefixes and tasks. Motivated by these findings, we introduce PathCal, a training-free controller that adjusts the strength of a fixed continuation bias using instantaneous competition among continuation, revision, and alternative-opening markers. It applies finite logit adjustments without additional model forward passes. An exact log-odds identity and frozen-prefix probes establish its local directional effect. Experiments show that PathCal improves the accuracy--length tradeoff in many mathematical reasoning settings and also reduces generation length in coding and multilingual mathematics.

cs.AI↗
arXiv2026

ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing

Recent video generation is increasingly realistic and controllable, yet video editing remains less developed, particularly for precise local edits that must preserve the original scene dynamics. Video scene text editing replaces text on scene surfaces, such as storefront signs, whiteboards, and product labels, while preserving the surrounding content, motion, and camera dynamics. Although scene text editing is well studied for images, video scene text editing that achieves high visual quality, temporal consistency, and edit locality remains underexplored. Existing resources offer limited paired real-video data, and general video-editing metrics do not directly measure whether the requested text remains correct over time. We introduce ViTeX-Bench, a benchmark suite comprising ViTeX-Dataset and a three-axis evaluation protocol. The dataset contains 387 real-world 720p videos with text-region masks and editing instructions: 230 provide reviewed, pipeline-generated paired edits for training, and 157 form a frozen evaluation split. The protocol evaluates text correctness, visual and temporal quality, and edit locality through 13 metrics, with one primary metric per axis and a Pareto comparison of their trade-offs. OCR calibration, human evaluation, and annotation-sensitivity analyses support the interpretation of these scores. Across eight baselines from four editing families, accurate text, temporal stability, and scene preservation remain difficult to achieve together. We also release ViTeX-Edit-14B, an open-source reference editor fine-tuned on the paired training split with motion-aligned glyph-video conditioning. It achieves CharAcc 0.688, the highest mean among the evaluated video-native editors, and the lowest comparable text-crop Warp among raw editor outputs. ViTeX-Bench provides a reproducible foundation for studying these trade-offs in video scene text editing.

cs.CV↗
arXiv2026

Training-Free Multi-Agent Cooperative Embodied Search from Uncertain Vision-Language Priors

Real-world Vision-Language Navigation (VLN) requires agents to act under partial observability, where map structure, semantic cues, and target locations may be incomplete or incorrectly inferred by foundation models. This paper presents a lightweight, training-free, and uncertainty-aware framework for cooperative multi-agent multi-target search in mapless environments. Instead of learning an end-to-end navigation policy, the framework separates high-level semantic reasoning from low-level execution through a scalable graph-optimization backbone. Multimodal Large Language Models (MLLMs) serve only as hypothesis generators, proposing semantic zones, target-location cues, and possible navigable connections. These outputs are treated as uncertain priors and further refined using online observations and graph-context evidence. Cooperative navigation is then formulated as a rolling-horizon multi-agent mixed-integer linear program with a lexicographic objective that first maximizes target-directed reward and then reduces travel distance, uncertain-edge use, and repeated visits. Experiments derived from 100 Matterport3D test cases show that the proposed framework improves target coverage and search efficiency over representative baseline methods, while shared information and joint planning enable more balanced workloads across agents. By operating on an abstract graph without relying on embodiment-specific policy parameters, the framework scales across varying numbers of agents and targets, generalizes to unseen environments, and accommodates different perception modules and low-level controllers without task-specific retraining.

cs.RO↗
arXiv2026

Generative AI for Autonomous Driving: Frontiers and Opportunities

Generative Artificial Intelligence (GenAI) constitutes a transformative technological wave that reconfigures industries through its unparalleled capabilities for content creation, reasoning, planning, and multimodal understanding. This revolutionary force offers the most promising path yet toward solving one of engineering's grandest challenges: achieving reliable, fully autonomous driving, particularly the pursuit of Level 5 autonomy. This survey delivers a comprehensive and critical synthesis of the emerging role of GenAI across the autonomous driving stack. We begin by distilling the principles and trade-offs of modern generative modeling, encompassing VAEs, GANs, Diffusion Models, and Large Language Models (LLMs). We then map their frontier applications in image, LiDAR, trajectory, occupancy, video generation as well as LLM-guided reasoning and decision making. We categorize practical applications, such as synthetic data workflows, end-to-end driving strategies, high-fidelity digital twin systems, smart transportation networks, and cross-domain transfer to embodied AI. We identify key obstacles and possibilities such as comprehensive generalization across rare cases, evaluation and safety checks, budget-limited implementation, regulatory compliance, ethical concerns, and environmental effects, while proposing research plans across theoretical assurances, trust metrics, transport integration, and socio-technical influence. By unifying these threads, the survey provides a forward-looking reference for researchers, engineers, and policymakers navigating the convergence of generative AI and advanced autonomous mobility. An actively maintained repository of cited works is available at https://github.com/taco-group/GenAI4AD.

cs.CV↗
arXiv2026

OpticalRec: Unified Optical Vision-Language Representation for Multimodal Recommendation

Recent advances in vision-language modeling have substantially improved multimodal encoding, retrieval and reasoning. Yet for multimodal recommendation, encoding rich item vision-language semantic interactions remains a long-standing bottleneck, which hampers accurate item representation learning and user-item matching. Mainstream approaches primarily adopt independent encoding of vision and language modality followed by rigid late fusion such as concatenation, inherently omitting native vision-language interactions and introducing cross-modal semantic distortion. To address this challenge, we propose OpticalRec, the first visual-space unified encoding paradigm for multimodal collaborative filtering, a fundamental recommendation setting. Instead of isolated modality-specific encoding, OpticalRec renders item textual metadata as visual glyphs, enabling native image-text interaction within the visual encoder - the perceptual encoding level. The resulting representations are further processed by the language decoder - the semantic encoding level, allowing OpticalRec to exploit the dual-attention mechanism of modern vision-language models that previous encoding methods omitted. OpticalRec's efficacy is theoretically supported by mutual information analysis and empirically demonstrated through superior performance across strong baselines and benchmarks. As a plug-and-play module, OpticalRec (1) introduces minimal cost, (2) is robust against rendered text font, color and layout, etc., and (3) integrates seamlessly into existing multimodal collaborative filtering models.

cs.IR↗
arXiv2026

The Missing Primitive: Diagnosing and Repairing Mathematical Reasoning in Large Language Models

While Large Language Models (LLMs) have demonstrated striking capabilities on frontier mathematical problems, it remains unclear whether they possess the structural mathematical understanding underlying their solutions. In this paper, we take a first step toward systematically studying mathematical understanding in LLMs, from diagnosing its distinct capabilities to leveraging these findings to improve post-training. First, we introduce the notion of Mathematical Primitive to probe structural mathematical understanding and propose \hlei{}, a novel benchmark that evaluates mathematical reasoning along four distinct dimensions: Discovery, Generation, Digestion, and Execution. Second, our systematic diagnosis shows that solution accuracy masks distinct capability profiles, primitives unlock substantial latent execution capacity, and Discovery is the dominant bottleneck in mathematical reasoning. Our post-training analysis further shows that discovery-limited failures are particularly amenable to repair. Finally, building on these findings, we introduce \abs{}, a primitive-privileged self-distillation framework that selectively transfers primitive-guided reasoning into the student model. Extensive experiments demonstrate that \abs{} consistently improves mathematical reasoning over baselines across model scales and challenging benchmarks.

cs.LG↗
arXiv2026

PISCO: Precise Video Instance Insertion with Sparse Control

The landscape of AI video generation is undergoing a pivotal shift: moving beyond general generation - which relies on exhaustive prompt-engineering and "cherry-picking" - towards fine-grained, controllable generation and high-fidelity post-processing. In professional AI-assisted filmmaking, it is crucial to perform precise, targeted modifications. A cornerstone of this transition is video instance insertion, which requires inserting a specific instance into existing footage while maintaining scene integrity. Unlike traditional video editing, this task demands several requirements: precise spatial-temporal placement, physically consistent scene interaction, and the faithful preservation of original dynamics - all achieved under minimal user effort. In this paper, we propose PISCO, a video diffusion model for precise video instance insertion with arbitrary sparse keyframe control. PISCO allows users to specify a single keyframe, start-and-end keyframes, or sparse keyframes at arbitrary timestamps, and automatically propagates object appearance, motion, and interaction. To address the severe distribution shift induced by sparse conditioning in pretrained video diffusion models, we introduce Variable-Information Guidance for robust conditioning and Distribution-Preserving Temporal Masking to stabilize temporal generation, together with geometry-aware conditioning for realistic scene adaptation. We further construct PISCO-Bench, a benchmark with verified instance annotations and paired clean background videos, and evaluate performance using both reference-based and reference-free perceptual metrics. Experiments demonstrate that PISCO consistently outperforms strong inpainting and video editing baselines under sparse control, and exhibits clear, monotonic performance improvements as additional control signals are provided. Project page: xiangbogaobarry.github.io/PISCO.

cs.CV↗
arXiv2026

ReSCENE: Server-Side Replay for Structural Mitigation of Catastrophic Forgetting in Federated Continual Learning

Federated continual learning must integrate new tasks over time without losing earlier-task knowledge. Most existing methods attach an anti-forgetting mechanism to the client-trained, server-aggregated loop of federated learning, which holds back new learning to preserve earlier knowledge and burdens resource-constrained clients. We propose ReSCENE, which structurally mitigates catastrophic forgetting by having each client upload a small condensed surrogate of its local data while the server keeps the surrogates of past tasks and trains the global model on them together with the current task surrogates. For efficient server memory, we introduce temporal herding, which selects the more recent surrogates from the pool accumulated over a task into a compressed buffer. Our study provides a theoretical analysis showing that this buffer can represent the original task data more closely than full accumulation of all surrogates. Across CIFAR-10, CIFAR-100, and TinyImageNet, ReSCENE achieves the strongest accuracy over seven baselines, by up to $31.1$ points of average accuracy, while requiring as little as $0.11\times$ of the client computation and up to $179\times$ less upload than the model-update baselines. ReSCENE further demonstrates its effectiveness when scaled to larger client populations and larger models while remaining efficient, which makes it a practical method for federated continual learning.

cs.LG↗
arXiv2026

Joint and Cross-Modal Video-Audio Generation and Editing: A Unified Formulation and Design Taxonomy

Video and audio are perceived together, yet most generative models treat them in isolation. We examine methods that model the two modalities jointly, generate one from the other, or edit them in a coupled manner, organized around a single question: how is the output kept coherent across modalities in time and semantics? A unified formulation casts joint generation, cross-modal generation, and joint editing as three problems defined on a single distribution over audio-visual pairs, and a taxonomy compares methods along five design axes. To our knowledge, this is the first overview to systematically taxonomize joint audio-visual editing, which we map as nine edit categories spanning 28 edit types. We describe methods, datasets, and metrics for each setting and close with the open problems we view as most consequential.

cs.CV↗
arXiv2026

CinematicVQA: Benchmarking Film-Grammar Reasoning in Large Vision-Language Models

Cinematography, the craft of visual storytelling through framing, lighting, and camera operation, fundamentally shapes how audiences perceive and emotionally engage with video content. While Large Vision Language Models (LVLMs) have made remarkable progress in video question answering, existing benchmarks primarily focus on identifying low-level techniques rather than understanding their storytelling impact. To address this, we introduce CinematicVQA, the first-of-its-kind benchmark for cinematic video understanding that goes beyond technique recognition to evaluate film-grammar reasoning, utilizing our introduced Cinematic Scene Graph (CSG), a structured representation that links filming techniques to their perceptual effects and narrative functions. Through comprehensive evaluation of state-of-the-art LVLMs, we reveal a striking semantic gap: models consistently perform higher on describing visual presentations than on identifying the underlying techniques. Surprisingly, Chain-of-Thought prompting fails to provide consistent gains and degrades performance for most models, suggesting that current LVLMs lack sufficient cinematic domain knowledge to benefit from step-by-step reasoning. Fine-tuning on \textsc{CinematicVQA-train} yields consistent improvements, particularly for narrative function and multi-hop reasoning. Overall, \textsc{CinematicVQA} serves both as a rigorous benchmark for cinematic evaluation in LVLMs and as a practical dataset for training more film-aware video models.

cs.CV↗
arXiv2026

AURORA: A Natural Language-Driven Agentic Framework for Understanding, Reasoning, and Orchestrating Reliable Air-Ground Co-Simulation

Air-ground transportation research increasingly relies on co-simulation, yet constructing scenarios remains labor-intensive and difficult to validate. More importantly, a generated scenario may execute successfully while failing to realize the spatial, temporal, communication, or behavioral relationships requested by the user. This paper presents AURORA, a natural-language-driven agentic framework that treats air-ground scenario generation as a process of compilation with verification. Central to AURORA is the Air-Ground Scenario Graph (AGSG), a typed intermediate representation that explicitly connects agents, aerial missions, events, communication links, success conditions, and their cross-domain dependencies. This shared representation enables simulator-grounded parsing, joint road-airspace grounding, temporal planning, pre-execution feasibility checking, trace-based runtime verification, failure localization, and bounded repair within a unified workflow. We further introduce AURORA-Bench to evaluate not only whether generated scenarios execute, but whether they faithfully realize the requested interactions. Experiments across multiple language models show that structured execution substantially improves reliability, while runtime verification exposes silent failures that completion-based evaluation overlooks. Localized repair further resolves many violations without regenerating the entire scenario. The results show that reliable scenario generation requires verifying realized behavior, not merely executable code, and demonstrate the value of explicit intermediate representations for verifiable and repairable language-driven co-simulation.

cs.RO↗